Breaking01 / AI
Emotional headlines raise AI severity ratings
Providing the article body reduced the effect of headline wording on AI severity ratings, the paper reported.

Researchers reported that emotionally intensified news headlines raised event severity ratings by 0.65–0.88 points in experiments with Chinese AI models.
The researchers published the paper on arXiv, a research preprint platform, on 11 Nov. According to its abstract, they analysed 63,926 Chinese news headlines and conducted experiments covering 30 real events, 30 synthetic events and 29 naturally occurring headline pairs.
The researchers reported that positively reframed headlines lowered severity scores by 0.34–0.52 points. Changing only the headline wording shifted assessments of both how misleading it was and how serious the event was.
The paper reported that supplying the article body reduced severity bias by 54–80%, compared with providing only the headline.
According to the abstract, the direction and ordering of the rating changes were the same across DeepSeek, MiMo, Dots and LongCat. A small human comparison study also found effects in the same direction.
The abstract does not disclose the range of the severity rating scale.